List your company
ResearchFormerly WizWand

Reward Modeling on Meta Safety (test)

64.5AccuracySafety RM
42.97248.56154.1559.739Jul 18, 2023
Updated 8mo ago

Evaluation results

MethodLinks
Safety RMFine-tuned=true2023.0764.5
Helpfulness RMFine-tuned=true2023.0762.8
GPT4zero-shot=true2023.0758.1
Open AssistantBackbone=DeBERTa V3 Large2023.0753.4
SteamSHP-XLBackbone=FLAN-T5-xl2023.0743.8